By DeVry University
August 31, 2020
44 min read
August 31, 2020
44 min read
Presenters:
In the medical field, artificial intelligence and machine learning are bringing a new set of tools to doctors and healthcare professionals. These resources aid in diagnosing patients, reading images, allocating hospital beds and designing applications with electronic health records.
In this Future-Ready Skills session, Dr. Shantanu Bose and Dr. Bob Arnot give us a glimpse into how these tools are being used to more effectively diagnose malignant melanoma and diabetic retinopathy while helping to increase the accuracy and efficiency of healthcare professionals.
So, let me give you an example. One of the neural networks is called a Convolutional Neural Network. And what that does is it analyzes images. And it's been the biggest breakthrough of all in terms of deep learning and artificial intelligence. When you look at these various scans that we would have, for instance, I think there’s something like two billion chest x-rays every year. So how could you learn from that? Well, what you do is you want to take a 100,000 of those x-rays for say lung cancer, and you take the readings that very, very good university professors have of those, so you truly know which ones are positive and which ones are negative. Then you have the Convolutional Neural Network go at it, and it’ll look at every little curve and every little snippet and be able to determine itself what's cancerous and what isn't cancerous.
And with that, then you have the ability to, in a small little hospital like we have up here in Vermont, take a CT scan, an MRI or a chest x-ray, and get as good a diagnosis as you would get at a top university medical center. So wonderful tools, incredibly helpful, very practical, and we see them all day long in everything we use.
Dr. Shantanu Bose: Right. That's great. Thank you for helping us understand those differences. Machine learning predominantly classified data into some segments and answering questions; yes and no type of questions ultimately. Deep learning the system really learns on its own once you give it a ton of data.
So that’s good advice.
Dr. Bob Arnot: If you don't like the professor, change. And just to give you a sense of this, many of these courses are like nine dollars. I took up a whole AI bootcamp that was many hours, and they're a very good overview. And I think I spent nine dollars on it. So, it isn't expensive, but it's just your time. Just think about your ordinary day. You don't have to sit down and plug away for an hour at a computer. You can literally walk around with your smartphone, plug in a chapter and get started.
I love the way you said that. And there is just a huge tech gap out there. If you look at jobs, the number one job in America now is data science in terms of salary. You have six-figure salaries for people early on in their careers. There's a huge need all across the spectrum, and it's wonderful, fun, and fascinating as a career. A lot of people, including my own kids, are looking at a pivot during this COVID era. They might have been in the service industry. I have an older son who has been into video production and whatnot, and he's going to do a big pivot just because now is an opportunity. If you're down on your luck, and a lot of us are, you have an opportunity to totally dig in, find something, pivot and really plan for a very bright future because this economy is going to come back and we are going to do well. But get ready for it. Don't sit around worried, anxious and depressed.
There's a phrase, "Don't let a crisis go to waste." And that's very much true. With this kind of a crisis, look around, look for opportunity. It may even be if you are unemployed, that it's a great opportunity for you be able to dig in, find an educational program, sign up, get up, get ready, as the economy comes back and be ready for a new life and a new era in economy 2.0. A lot of people think that economy 1.0 as a vast service economy, wasn’t that great and that 2.0, with the ability to work anywhere, use these Zooms, have all the amazing tool sets on your computer, frees us to live where we want and get much more involved in the information and artificial intelligence era and something we love and that has real legs as a career.
One of the things that is so fun about machine learning as you start out, is there are these wonderful libraries. Think of a library as like taking an app off an iPhone or smartphone – and one of them is called Scikit-Learn. With these libraries, you can just graph out the relationships between the data.
So as an example, let's say, you're trying to determine whether or not somebody has congestive heart failure. And in your dataset, you have a thousand people without congestive heart failure, and you have one person with congestive heart failure. Well, it's terribly unbalanced data. There's no way you can get anything out of it. So, you're sizing up the data. Does this make sense? Do we think we have relationships in this data?
And what you'll find is you take just two factors, and that might be, say, blood pressure and blood sugar for diabetes. With heart disease, there’s a pretty tight connection a tight connection between sex and age. You know, a pretty tight connection there. But you're looking for more. So, you may have a hundred columns there and you'll see little relationships there, but that's where the magic of machine learning and deep learning comes in. When people ask about machine learning, I take this example. Shantanu, your parents probably did this. You go out to breakfast and it's like, "Oh my God, I have to spend breakfast with my parents again. What am I going to do?" Well, they bring out this little sheet of paper that has a hundred dots on it. You know, and you go “I wonder what that is.” You start to fill them in and as you're filling it in, at the 70th dot out of a hundred, you go, “That's a bear.”
So, what machine learning does is it’s able to recognize patterns. If there's one concept to walk away with it's this. When I started computer programming, I'm sure when you started coding too, you know, these machine level program, incredibly complex to learn to write thousands of lines of code to get anything out. And this was completely beyond the reach of anybody. The real joy of machine learning is data is writing the computer program for you. So, you take really good, clean, wonderful data, you put it in there, and the machine is writing what we call an algorithm. It’s figuring all the stuff out, it’s writing out an algorithm. So once it's trained up, then you could take your data and put it in.
As an example, let's say you have a prediction of whether somebody has diabetes or not. So you have a training set and that training set basically teaches the computer. The data goes in there and the computer looks at it and molds it. Oh yeah, I see those relationships. And then you would test it. Does this really hold up? Look at the test data, look at the accuracy – being 70, 89, 95, 98%. And then finally, you're ready to use your own data. It's ready to go into action. It's a joy to think you get to just you dump it in and you're sure it's good data. You know what you're looking for there, and it writes the program for you, which is what's transformative over here because coding is hard.
So good question there.
Filter Blog Post Category